Online Detection and Mitigation of Robust Zero Dynamics Anomaly Behavior in MIMO Nonlinear Control Systems

TL;DR

This paper introduces an online detection and mitigation approach for zero dynamics anomalies in MIMO nonlinear control systems, ensuring stability without precise models by using residual signals and neural networks.

Contribution

It proposes a model-free, two-stage method combining residual-based detection and neural network recovery for robust anomaly mitigation in nonlinear systems.

Findings

Effective anomaly detection via residual signals

Neural network-based recovery maintains system stability

Validated on a four-tank system simulation

Abstract

This paper presents a methodology to detect robust zero dynamics anomaly behavior and mitigate the impacts in general multi-input multi-output (MIMO) nonlinear systems. The proposed method guarantees the resiliency and stability of the closed-loop system without relying on an accurate dynamical model. The presented method operates in two stages. First, it measures the difference between the system input and that of the model as a residual signal to detect the anomaly behavior. After detecting the attack, a recovery signal is generated to restore the system to its nominal condition. In this stage, a neural network model is used to estimate the anomaly signal and recover the closed-loop system. The weights of the neural network model are updated online using adaptation rules without needing prior data for training. The accuracy and performance of the proposed methods are verified by simulating various scenarios on a fourtank system.